Puzzle 482
Puzzle 482 Advanced Alignment Puzzle irc_108474_1322507548 E1 Contenders Bletchley Park.png|E1 contenders Bletchley Park irc_108474_1322507908 E1 contenders Bletchey Park.png|E1 Contenders Bletchley park irc_202811_1322206880 S4 ch garnier.png|S4 Ch Garnier irc_202811_1322206855 S4 ch garnier.png|S4 Ch Garnier irc_174969_1322180963 S7 rav.png|S7 Rav3n_pl irc_174969_1322180938 S7 rav.png|S7 Rav3n_pl irc_174969_1322180917 S7 rav.png|S7 rav3n_pl irc_235191_1322190028 s14 gramps.png|S14 gramps irc_328083_1322190046 s68 wildcat.png|S68 Wildcat Methods Used by the Contenders Team for this Puzzle 1. First, infjamc looked up the sequence on the RCSB protein data bank. The idea is to see if there are other known structures that could use as off-game templates, to get a "second opinion" on the secondary structure predictions. To his surprise, there was an exact hit (2L3B), but it's an NMR structure. From this finding, he figured that an exact X-ray crystal structure is most likely either not yet known or at least not yet published. 2. The PDB file of the NMR model was downloaded for further processing. In PyMol, the residues not found in the puzzle were removed, and this processed version of Model 1 was then loaded in the stand-alone version of Foldit (which is used by researchers) for an estimate of the score of the native structure. A quick shake and wiggle yielded an energy score around -280, which translates to a Foldit score of 10280. 3. Then, the secondary structure "prediction" of the NMR structure was manually entered into the Foldit game. The general topology (one beta sheet on each side) of the protein, as well as the hydrogen bonding pattern of the beta sheet, were also copied manually. To make this process easier, an alignment was applied to improve the match in the secondary structure predictions with the 2L3B template (with the gaps given to the loops whenever possible). 4. At this point, infjamc decided to extract no further information from the 2L3B "template" for two reasons: (a) out of concern of fairness, and (b) because NMR structures tend to be less reliable in their loop sections relative to X-ray structures anyway. Instead, the protein was processed "normally" from that point on as if it were any other puzzle. For example, a few rebuilds were conducted, which led to a few minor changes in the secondary structure settings around the edges of the beta sheets. 5. After that, infjamc's initial attempt was uploaded, with a score around 10085 (later improved to 10180). This solution was further processed by various other members on the Contenders team, including Bletchley Park, BootsMcGraw, CharlieFortsConscience, Mark, mimi, and spvincent. This first round of evolving, which mostly involved rebuilding via scripts, resulted in three solutions that scored in the 103xx range. 6. Upon seeing his teammates' solutions, infjamc attempted a 2-way merge by loading the highest-scoring team solution, with the second-highest-scoring solution loaded as as guide. Scoring relative to guide was turned on to allow for ease of mixing-and-matching the best aspects of both solutions. When needed, two clients would be open at the same time for a direct side-by-side comparison and manual copying, and the Alignment Tool was used from time to time for copying an entire loop. The same was repeated for the 3rd highest scoring solution, and the result was a hybrid that scored in the 10400s. He then attempted to reverse-engineer the resulting hybrid in soloist mode for an even higher score (after conducting several side chain tweaks in the full-atom viewing mode) and then re-uploaded the result to the team. 7. Steps #5 and #6 were repeated for several more times over the next few days. To the surprise of everyone on the Contenders team, the aggregate result was that more than 500 points were gained from loop rebuilds alone. (This was confirmed by loading infjamc's earlier 10180 soution as a guide-- the beta sheet region was virtually unaffected despite the many rebuilds, while the loop sections deviated significantly.) 8. The last hybrid solution that infjamc managed to upload before the puzzle closed scored 10658, which Bletchley Park processed via a continuous rebuilding script. This resulted in the 10749 score that registered on the server, although further processing after the closing of the puzzle ultimately led to the solution that scored 10776 in the screenshot. Bottom line: The high score was the result of a good starting structure plus extensive teamwork that utilized the different strengths of the various members in the Contenders group. Note : Infjamc from the Contenders was kind enough to give me an answer when I asked him about his work that led to his being the best solution for puzzle #482. He also gave me permission to repost his response to me so that everyone could share in the educational aspect of game. I found his answer especially gracious since I'm a member of the Go Science team. - smilingone Category:Puzzle Results Category:Groups Category:Beta Category:Strategy